Checking
Data and scoring
Healthy freshness and tables do not prove investment performance.
MLOps
MLOps
Review model quality, score gates, and promotion readiness.
Model Validation
Shadow/Paper gate
Review shadow evidence and paper gate lock status, not an executable service.
Performance Compare
Benchmark Compare
Compare operating strategy PnL against BTC, ETH, and SOL benchmarks over the same period.
Model Operations
Model and Data Status
Check whether scoring, table freshness, and active model status are healthy.
Quant operations decision
Data is running, but cost-adjusted alpha is not proven
Read net return, mature samples, calibration, and execution gates before model count or classification scores.
Not proven
Cost-adjusted alpha
The current entry baseline and directional challengers do not clear promotion evidence.
Blocked
Paper / Live execution
Closed until stable candidates, cost-adjusted shadow evidence, and explicit human approval exist.
Model lifecycle
Keep only two active research questions
Separate risk-support models from alpha candidates and hide rejected or superseded models by default.
Regime v4
QUIET · RANGE · VOLATILE risk state
Keep as read-only exposure context, never direction or alpha.
GARCH volatility
Volatility risk context
Not a buy/sell signal. Review artifact and training-window freshness separately.
Directional opportunity h5/h15/h60
One directional experiment with three horizons
Repair point-in-time regime coverage and pass cost-adjusted walk-forward evaluation.
Legacy liquidation entry filter
Sparse entry-filter research
Current PF and average bps lack a cost buffer; collect a larger mature sample.
Entry baseline
Negative-performance comparison control
Not candidate alpha; retain only to expose overconfident p_good calibration.
6 stopped or archived models
Regime supervised v2
Rejected after directional labels failed forward-return alignment.
Regime supervised v3
Comparison artifact superseded by nondirectional v4.
Entry quality logreg v2
No paper-quality candidate; retain lineage only.
Entry logreg/LGB v1
Failed exact-cost and walk-forward gates.
SL/TP RF v1
No current approved execution role or promotion evidence.
EV-v3 · DOGE label variants
Threshold tuning did not recover repeatable cost-adjusted edge.
This is an operating decision contract. Stop PM2 processes only after mapping each process input, output, consumer, and 24-hour impact.
1. Data and label healthReview freshness, missing data, point-in-time coverage, and label maturity before models.
Model Data Status
Quickly shows whether model scores and data freshness are ready.
Connection
Checking
Last check
Checking
Data freshness
Checking data status
Latest score
No data
Score lag
Checking
Latest data
No data
Data Freshness
Shows how recently the data required for model decisions was updated.
TableLatest dataLatest updateLagRowsStatus
Checking data status.
2. Alpha validationUse net bps, profit factor, mature samples, and calibration rather than AUC alone.
Signal Quality
Samples showing whether candidate signals were correct after maturity. If this is weak, fix the model/features before backtests.
Use win rate above 55% as the first watch criterion.
Check whether avg return is positive after costs.
Exclude small samples from promotion decisions.
Model Quality Triage
Operational judgment to check before backtest results. Confirms whether the model can create trades or should stay in risk-blocking/Shadow observation only.
Checking model quality
Deployment
Checking
Check operation mode and risk limits together.
Score freshness
No recent score
The runtime gate cannot create entry candidates without a recent score.
Active samples
Checking
Model validation decisions are empty without an active scoring model.
Data trust
Checking
Input data freshness gate is passing.
3. Execution gateSeparate Shadow observation from Paper/Live promotion and preserve human approval.
Promotion Review
Minimum requirements before a model affects paper/live gates. Items that cannot be verified with current data are not treated as passing.
Registered / active model
The operation runtime must reference an actual model version.
No model version or active score model is available.
Recent score gate
Recent scores must be able to pass the runtime threshold.
Recent score is missing or stale.
Sample size
Minimum samples are required so symbol/side decisions are not driven by chance.
active rows 24h=0 · minimum 100
Data freshness
Model quality judgment is paused if any feature, signal, or score table is stale.
No table freshness issues.
Paper/live parity
Confirm that Shadow results and Paper/live execution quality point in the same direction.
Live parity has no separate evidence yet. This gate uses shadow trades.
Promotion mode
Apply risk blocking or position sizing adjustments before entry generation.
Operation mode also needs limits and a kill switch.
Execution Readiness
Recent 1-hour view used by the MLOps trader. Unlike the 24-hour decision mix, this value must pass before an entry event is created.
BTCLong
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
BTCShort
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
ETHLong
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
ETHShort
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
SOLLong
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
SOLShort
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
DOGELong
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
DOGEShort
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
TRXLong
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
TRXShort
No recent scoreRows
Checking
Pass
Checking
Max / req
Checking
Gap
Checking
Latest score: None
Shadow Trading Events
The records below are shadow trading only. This screen does not imply live execution.
MLOps shadow observation events
Last 24h · shadow evidence for promotion review
Entries 0Exits 0Failures 0
This panel shows MLOps shadow evidence only. PASS_CANDIDATE is not a trading signal, and recent activity can look old while promotion is locked.
MLOps entry notional
₩0
MLOps exit notional
₩0
Realized P&L
₩0
Open operational control-room details
Advanced diagnostics · raw model and score details
Model Status
Checking model operations
Checking operation mode. Checking model scores and data freshness.
Checking
Latest model
Checking
Registry model Checking
Active Models
Checking
Checking status
Last check
Checking
Run schedule
Score Checking
Train Checking KST
24h scores
Checking
Checking status
Latest scoring
No data
Latest model score creation time
Active Models
Checking
Checking status
Run schedule
Checking
Train Checking KST · Checking operation mode
Active Models
Models used for recent score calculation.
Checking active models.
Decision Mix
Shows which direction model decisions took over the last 24 hours.
Checking decision mix.
Market Data Guide